### Install VIP Package

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Installs the VIP package in editable mode from the current repository directory.

```bash
pip install -e .
```

--------------------------------

### Run Training for Locomotion Environments

Source: https://github.com/clvrai/furniture-bench/blob/main/implicit_q_learning/README.md

Starts the training process for locomotion environments like HalfCheetah. Requires a specified environment name and configuration file.

```bash
python train_offline.py --env_name=halfcheetah-medium-expert-v2 --config=configs/mujoco_config.py
```

--------------------------------

### Install System Dependencies

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Install necessary system packages for MuJoCo and rendering. This includes packages for CMake, MPI, OpenGL, and GLEW.

```bash
$ sudo apt-get install cmake libopenmpi-dev libgl1-mesa-dev libgl1-mesa-glx libosmesa6-dev patchelf libglew-dev

# software rendering
$ sudo apt-get install libgl1-mesa-glx libosmesa6 patchelf

# window rendering
$ sudo apt-get install libglfw3 libglew-dev
```

--------------------------------

### Basic CMake and Catkin Setup

Source: https://github.com/clvrai/furniture-bench/blob/main/furniture_bench/assets/franka_description_ros/franka_description/CMakeLists.txt

Sets the minimum CMake version, defines the project name, and finds the catkin package. It also declares catkin dependencies.

```cmake
cmake_minimum_required(VERSION 3.4)
project(franka_description)

find_package(catkin REQUIRED)
catkin_package(CATKIN_DEPENDS xacro)
```

--------------------------------

### Install Directories

Source: https://github.com/clvrai/furniture-bench/blob/main/furniture_bench/assets/franka_description_ros/franka_description/CMakeLists.txt

Installs the 'meshes' and 'robots' directories to the catkin package's share destination.

```cmake
install(DIRECTORY meshes
  DESTINATION ${CATKIN_PACKAGE_SHARE_DESTINATION}
)
install(DIRECTORY robots
  DESTINATION ${CATKIN_PACKAGE_SHARE_DESTINATION}
)
```

--------------------------------

### Install RoLF Package

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Install the Robot Learning Framework (RoLF) package in editable mode. This command should be run from the root directory of the project.

```bash
# at the root directory (`robot_learning/`)
$ pip install -e .
```

--------------------------------

### Install Dependencies for Implicit Q-Learning

Source: https://github.com/clvrai/furniture-bench/blob/main/implicit_q_learning/README.md

Installs necessary Python packages, including JAX with CUDA support. Ensure your CUDA and cuDNN versions are compatible.

```bash
pip install --upgrade pip

pip install -r requirements.txt

# Installs the wheel compatible with Cuda 11 and cudnn 8.
pip install --upgrade "jax[cuda]>=0.2.27" -f https://storage.googleapis.com/jax-releases/jax_releases.html
```

--------------------------------

### Run Finetuning on AntMaze Tasks

Source: https://github.com/clvrai/furniture-bench/blob/main/implicit_q_learning/README.md

Starts a finetuning process on AntMaze tasks. Allows for custom evaluation parameters and replay buffer size. Requires specifying the environment and configuration.

```bash
python train_finetune.py --env_name=antmaze-large-play-v0 --config=configs/antmaze_finetune_config.py --eval_episodes=100 --eval_interval=100000 --replay_buffer_size 2000000
```

--------------------------------

### Run GAIL with BC Initialization

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Execute GAIL with initialization from a BC checkpoint. This allows GAIL to start with a pre-trained policy from Behavior Cloning. Specify paths for demo data and the BC checkpoint.

```bash
# GAIL with BC initialization
python -m rolf.main run_prefix=test algo@rolf=gail env.id=Hopper-v2 demo_path=log/Hopper-v2.ppo.test.123/demo/Hopper-v2.ppo.test.123_step_00001000000_100.pkl init_ckpt_path=log/Hopper-v2.bc.test.123/ckpt_00000020.pt init_ckpt_pretrained=True
```

--------------------------------

### Create and Activate Conda Environment

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Sets up a new conda environment named 'vip' with Python 3.9 and activates it for package installation.

```bash
conda create --name vip python=3.9
conda activate vip
```

--------------------------------

### Install PyTorch with CUDA 11.3

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Install a specific version of PyTorch (1.10.2) compatible with CUDA 11.3. This command specifies the PyTorch, torchvision, and torchaudio versions.

```bash
# PyTorch 1.10.2, Linux, CUDA 11.3
$ pip3 install torch==1.10.2+cu113 torchvision==0.11.3+cu113 torchaudio==0.10.2+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html
```

--------------------------------

### Install MuJoCo 2.1.0 and 2.1.1

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Download and extract MuJoCo versions 2.1.0 and 2.1.1. Ensure the correct versions are placed in the ~/.mujoco directory and the LD_LIBRARY_PATH is updated for MuJoCo 2.1.0.

```bash
# download MuJoCo 2.1.0 for mujoco-py
$ mkdir ~/.mujoco
$ wget https://github.com/deepmind/mujoco/releases/download/2.1.0/mujoco210-linux-x86_64.tar.gz -O mujoco210_linux.tar.gz
$ tar -xvzf mujoco210_linux.tar.gz -C ~/.mujoco/
$ rm mujoco210_linux.tar.gz

# download MuJoCo 2.1.1 for dm_control
$ wget https://github.com/deepmind/mujoco/releases/download/2.1.1/mujoco-2.1.1-linux-x86_64.tar.gz -O mujoco211_linux.tar.gz
$ tar -xvzf mujoco211_linux.tar.gz -C ~/.mujoco/
$ rm mujoco211_linux.tar.gz

# add MuJoCo 2.1.0 to LD_LIBRARY_PATH
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/.mujoco/mujoco210/bin

# for GPU rendering
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/nvidia
```

--------------------------------

### Generate Embedding Distance Curves

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Navigates to the examples directory and runs a script to generate embedding distance curves. This script produces plots in the 'embedding_curves' subdirectory.

```bash
cd vip/examples
python plot_reward_curves.py
```

--------------------------------

### Run Training for Kitchen and Adroit Environments

Source: https://github.com/clvrai/furniture-bench/blob/main/implicit_q_learning/README.md

Executes the training script for Kitchen and Adroit environments. Requires the environment name and its corresponding configuration file.

```bash
python train_offline.py --env_name=pen-human-v0 --config=configs/kitchen_config.py
```

--------------------------------

### Run PPO Algorithm

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Use this command to run the Proximal Policy Optimization (PPO) algorithm. Specify the run prefix, algorithm, and environment ID.

```bash
python -m rolf.main run_prefix=test algo@rolf=ppo env.id=Hopper-v2
```

--------------------------------

### Train VIP on Custom Video Dataset

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Initiates VIP training on a custom video dataset. Specify the dataset name and path using the provided arguments.

```bash
python train_vip.py --config-name=config_vip dataset=my_dataset_name datapath=my_dataset_path
```

--------------------------------

### Run Training for AntMaze Environments

Source: https://github.com/clvrai/furniture-bench/blob/main/implicit_q_learning/README.md

Initiates training for AntMaze environments. Allows for custom evaluation episodes and intervals. Requires specifying the environment name and configuration.

```bash
python train_offline.py --env_name=antmaze-large-play-v0 --config=configs/antmaze_config.py --eval_episodes=100 --eval_interval=100000
```

--------------------------------

### Run Behavior Cloning (BC)

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Execute the Behavior Cloning algorithm using the collected expert trajectories. Specify the run prefix, algorithm, environment ID, and the path to the demonstration data.

```bash
python -m rolf.main run_prefix=test algo@rolf=bc env.id=Hopper-v2 demo_path=log/Hopper-v2.ppo.test.123/demo/Hopper-v2.ppo.test.123_step_00001000000_100.pkl
```

--------------------------------

### Run GAIL Algorithm

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Use this command to run the Generative Adversarial Imitation Learning (GAIL) algorithm. Specify the run prefix, algorithm, environment ID, and the path to the demonstration data.

```bash
python -m rolf.main run_prefix=test algo@rolf=gail env.id=Hopper-v2 demo_path=log/Hopper-v2.ppo.test.123/demo/Hopper-v2.ppo.test.123_step_00001000000_100.pkl
```

--------------------------------

### Train VIP on Ego4D Dataset

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Trains VIP using the 'config_vip_ego4d.yaml' configuration, suitable for large-scale pre-training on the Ego4D dataset. Provide the path to your Ego4D dataset.

```bash
python train_vip.py --config-name=config_vip_ego4d dataset=ego4d datapath=ego4d_dataset_path
```

--------------------------------

### Integrate VIP with TorchRL Environment

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Applies the VIPTransform to a TorchRL environment to use VIP as an out-of-box visual representation. Ensure 'my_env' is an existing environment and 'next_pixels' is the key for pixel observations.

```python
from torchrl.envs.transforms import VIPTransform
env = TransformedEnv(my_env, VIPTransform(keys_in=["next_pixels"], download=True))
```

--------------------------------

### Run DDPG Algorithm

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Use this command to run the Deep Deterministic Policy Gradient (DDPG) algorithm. Specify the run prefix, algorithm, and environment ID.

```bash
python -m rolf.main run_prefix=test algo@rolf=ddpg env.id=Hopper-v2
```

--------------------------------

### Run TD3 Algorithm

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Use this command to run the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. Specify the run prefix, algorithm, and environment ID.

```bash
python -m rolf.main run_prefix=test algo@rolf=td3 env.id=Hopper-v2
```

--------------------------------

### Run SAC Algorithm

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

Use this command to run the Soft Actor-Critic (SAC) algorithm. Specify the run prefix, algorithm, and environment ID.

```bash
python -m rolf.main run_prefix=test algo@rolf=sac env.id=Hopper-v2
```

--------------------------------

### Collect Expert Trajectories for BC

Source: https://github.com/clvrai/furniture-bench/blob/main/rolf/README.md

First, train a PPO expert agent. Then, collect expert trajectories using the trained policy. These trajectories are stored for use in Behavior Cloning.

```bash
# train ppo expert agent
python -m rolf.main run_prefix=test algo@rolf=ppo env.id=Hopper-v2

# collect expert trajectories using ppo expert policy
python -m rolf.main run_prefix=test algo@rolf=ppo env.id=Hopper-v2 is_train=False record_video=False record_demo=True num_eval=100
```

--------------------------------

### Load Pre-trained VIP Model

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Loads the VIP model pre-trained on Ego4D and sets it to evaluation mode.

```python
from vip import load_vip
vip = load_vip()
vip.eval()
```

--------------------------------

### Train R3M Representation

Source: https://github.com/clvrai/furniture-bench/blob/main/r3m/README.md

Command to train the R3M representation using the Ego4D dataset. Requires specifying data paths and Weights & Biases project details.

```bash
python train_representation.py hydra/launcher=local hydra/output=local agent.langweight=1.0 agent.size=50 experiment=r3m_test dataset=ego4d doaug=rctraj agent.l1weight=0.00001 batch_size=16 datapath=<PATH TO PARSED Ego4D DATA> wandbuser=<WEIGHTS AND BIASES USER> wandbproject=<WEIGHTS AND BIASES PROJECT>
```

--------------------------------

### Load R3M Model

Source: https://github.com/clvrai/furniture-bench/blob/main/r3m/README.md

Load a pre-trained R3M model for use in your code. Supports different model sizes like resnet50 and resnet18.

```python
from r3m import load_r3m
r3m = load_r3m("resnet50") # resnet18, resnet34
r3m.eval()
```

--------------------------------

### Add Nosetests for Testing

Source: https://github.com/clvrai/furniture-bench/blob/main/furniture_bench/assets/franka_description_ros/franka_description/CMakeLists.txt

Conditionally adds Python nosetests for robot URDF files if testing is enabled in catkin.

```cmake
if(CATKIN_ENABLE_TESTING)
  catkin_add_nosetests(test/franka_robot_urdf.py)
  catkin_add_nosetests(test/hand_urdf.py)
  catkin_add_nosetests(test/dual_panda_example_urdf.py)
endif()
```

--------------------------------

### Comment out NaN/Inf checks in Gym

Source: https://github.com/clvrai/furniture-bench/blob/main/implicit_q_learning/README.md

Modifies the `passive_env_checker.py` file to comment out checks for NaN and infinity values in observations. This is a workaround for specific environment issues.

```python
    # if np.any(np.isnan(obs)):
    #     logger.warn("Encountered NaN value in observations.")
    # if np.any(np.isinf(obs)):
    #     logger.warn("Encountered inf value in observations.")
```

--------------------------------

### FurnitureBench Citation

Source: https://github.com/clvrai/furniture-bench/blob/main/README.md

Use this BibTeX entry to cite the FurnitureBench paper in your research.

```bibtex
@inproceedings{heo2023furniturebench,
title={FurnitureBench: Reproducible Real-World Benchmark for Long-Horizon Complex Manipulation},
author={Minho Heo and Youngwoon Lee and Doohyun Lee and Joseph J. Lim},
booktitle={Robotics: Science and Systems},
year={2023}
}
```

--------------------------------

### VIP Citation

Source: https://github.com/clvrai/furniture-bench/blob/main/vip/README.md

Provides the BibTeX entry for citing the VIP paper in academic research.

```bibtex
@article{ma2022vip,
  title={VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training},
  author={Ma, Yecheng Jason and Sodhani, Shagun and Jayaraman, Dinesh and Bastani, Osbert and Kumar, Vikash and Zhang, Amy},
  journal={arXiv preprint arXiv:2210.00030},
  year={2022}
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.